Seoul adopts AI for suicide prevention on Han Bridge: Uni boffins train machine learning model on rescue teams' data

System trains itself to spot suicidal behavior and alert a controller

South Korea's capital and largest metropolis has turned to AI for suicide detection and prevention on popular bridges along the prominent Han River, according to the system's developers at the Seoul Institute of Technology (SIT).

There are 27 bridges that cross the Han River, also known as the Hangang river, in the Seoul National Capital Area (​divided into Seoul, Gyeonggi, Incheon). Many of the city's pedestrians walk on them every day. Unfortunately, only counting suicide attempts from Seoul bridges, researchers said there are an average of 486 people trying to end their lives in the Han's waters every year. This translates to a large amount of required rescue resources, which thankfully are quite successful, at 96 per cent. A more efficient spread of resources, however, would mean being able to potentially save more lives.

With the current setup, which according to an SIT document went live in February, the bridges are monitored by an array of CCTV screens at a control centre, with each small box depicting a different part of the bridge. An employee watches these screens for unusual behaviour that may require interference or a rescue worker. The AI collaboration between SIT and Seoul Fire and Disaster Headquarters is aimed at helping the technicians to better shift their focus by using machine learning to alert them to the scenes most likely to need intervention.

"The objective is to establish an effective control system and reduce the casualties caused by falling accidents," according to the SIT document, when translated to English via automated language tools.

The collaboration between the uni boffins and the workers from Seoul Fire and Disaster HQ has been providing data since 2020, detecting and predicting situations by learning behavioural patterns of those needing assistance and showing those images on the controller’s monitor. These patterns could consist of actions like pulling on cables as detected by sensors. The hope is that ultimately the system learns "learning the [behaviour] pattern of the attempter" and rescuers have more time to act before the individual takes action.

The system continually learns, analyzing structured and unstructured data such as report history and text of call details, as well as sensor data, to reduce false alarms and improve accuracy. The characteristics of the bridge and changes of weather, such as light and wind, are also accounted for.

A video posted on SIT’s Facebook page and YouTube channel shows an engineer helming the monitor system. In a scene watched by the engineer, the area outside the bridge is greyed out and not in focus, and the movement caused by wind is stabilised.

Youtube Video

The engineers on the video said the system minimises the monitoring gap, allowing quicker and more accurate response to incidents, thus saving lives. Currently the rescue rate is around 96 per cent.

Korea has the highest suicide rate of any country in the OECD at 23 per 100,000 people between 2016 and 2018. COVID hardships have since increased these numbers and the number of Han River bridge rescue dispatches, reports Reuters. ®

TalkNote

To those reading this who have been having some not-so-good thoughts, click here to find a number to talk to someone. Things will look different tomorrow.